Associative Memory

1
Associative Memory

• Sang Ki Park

2
Table of Contents

• Associative Memory Definition
• Memory Storing Recall
• Simple Pattern Experiment
• Character Recognition

3
Associative Memory

• When the incomplete or distorted pattern is
  given, Associative Memory analogy the original
  pattern.
• Hopfield network is an effective associative
  memory.

4
Memory Store Recall
Memory Recall
Memory Store
5
Associative Memory Process
A B C D
A B C D
A B C D
A B C D
6
Simple Pattern Experiment

• Stored Pattern
• S 1 1 1 1 0 0 0 0, 0 0 0 0 1 1 1 1
• Input 1 1 1 1 0 0 0 0 ? Output 1 1 1 1 0 0 0
  0
• Input 0 0 1 1 0 0 0 0 ? Output 1 1 1 1 0 0 0
  0
• Input 0 0 0 0 1 1 0 0 ? Output 0 0 0 0 1 1 1
  1
• Input 0 1 0 0 0 0 1 1 ? Output 0 0 0 0 1 1 1
  1

Associative Memory Network can recall original
pattern from distorted input.
7
Character Recognition

• Store Character Patterns and Recall them

8
Training
Noise 0
Noise 10
Noise 30
Noise 40
Noise 20
Input
Output
Associative Memory Network has a robustness about
Noise or distorted Input
9
Pattern similarity Problem

• Patterns are too similar to be classified

?
D
B
?
O
C
?
10
Similarity Measure

• Hamming Distance

11
Training
Noise 20
Noise 0
Noise 10
Input
Output
Stored Similar Patterns confuses associative
memory
12
Reducing the Patterns Similarity

• Transform the pattern similar to other patterns

B C D O
O
O
13
Similarity Measure

• Hamming Distance

O Transformed O
14
Training
Noise 10
Noise 20
Noise 30
Noise 0
Input
Output
After reducing the patterns similarity,
Recognition performance get better Therefore, It
is needed to increase the difference between
training patterns
15
Conclusion

• Associative Memory Network is a good method to
  recall Original pattern from noise or distorted
  input pattern
• Storing Pattern should not consist of too similar
  patterns
• Need to consider how to handle similar patterns